So anyway, it moves the Asset downloading (packet sending) to a module (AssetDownloadModule).
So now at last, AssetCache should be just dealing with fetching assets from the asset server and caching them.
* This might stop some client's constant requests for unfound textures, which is a candidate for the memory leak
* If a texture is not found then the "Image not found" texture will now be displayed clientside
* If it works, this should resolve mantis 676
* Non texture image requests do not receive this packet yet
* This will require a prebuild
* As such, only a request for a non cached asset, the response and failures show up now.
* I know lbsa71 only put these in not long ago, so if they are really still required, I think we should think whether we can move the default log4net level off 'Debug'
* This will show the packets waiting in each queue for each client logged into a region server
* These are displayed using 'show stats' on the region command line
* This is in pursuit of a memory leak.
* This will require a prebuild
* Resolve a bad logic bug in AssetCache.GetAsset()
* This may make some asset related things work better (possibly getting main map images will now be improved).
* This should stop the constant increase in the download requests statistics
* If you see stat numbers for download requests which are far from what you'd expect, please report
* This fixes some of the 'runaway downloads' problem but not all of it
* Also fix up logging messages so texture requests are reported as such rather than as assets
Patch provided by Openlifegrid.com
Adds locks around clientCircuits and clientCircuits_reverse Adds existance check on adding to clientCircuits for clients resending UseCircuit packet.
Adds locks around Clientmanager.m_clients for add/remove/tryGet calls
* This will make it more obvious if one accidentally starts one's region server in standalone rather than grid mode but then tries to login to a grid (as I am wont to do)
* We are now sending back a fault code (-32601) if no xmlrpc method is found rather than an ordinary message
* Converted all instances of "256" spotted to use RegionSize instead. Some approximations used for border crossings (ie 255.9f) are still using that value, but should be updated to use something based on RegionSize.
* Moving Terrain to a RegionModule, implemented ITerrainChannel and TerrainModule - nonfunctional, but will be soon.
* Made regions table name configurable (MSSQL only)
* Added a note in ini.example pointing out that the sql resources have to change if you change table names
* Removed duplicate picker method from GridData interface
[Provided by openlifegrid.com]
* Push asset update (invoked when clothing is altered) down into AgentAssetTransactions from Scene.Inventory.cs to join others
* I've tested that clothing creation and update still works, but please let me know if it suddently breaks for you.
* Add/correct comments
these two patches fixe the terrain load exception that occurs when
restarting OpenSim on PPC32 by making terrain loading endian clean.
it does NOT fix the client crash exception. need to hunt that one down.
'show assets' shows the current state of the asset cache (number of cached assets, requests, et c)
'clear-assets' forcibly re-initializes the asset cache thereby freeing all cached items.
'clear-assets' is not to be used lightly, as it probably introduces mem inconsistencies and doubling up of textures.
* Made Estate tools work for estate managers without needing to request admin status First
* Added code to make the Simulator version to be reported in the About box of the client
* Still not enough to solve the memory leak, though hopefully this is another step on the path
* All these changes are pretty temporary - this will be addressed with a more fundamental refactor in the future
* ParcelProperties are sent only when needed instead of on any movement
* Terse Updates and other temporary data packets are marked unreliable
* After a certain amount of users, the sim actually sends updates on things less
* Experimental
* Tested to 68 avatar with pCampBot (And it's surprising what actually causes the most lag.. the text chat!)
* The immediate upshot is that "show uptime" from the console will now show uptime on all server types (user, asset, grid, etc)
* DEV: This refactoring is far from complete - only just enough to makes the "show uptime" command common accross the servers. More is needed, but in this case it's somewhat like eating cabbage, which I prefer not to do all at once
* This may help with asset server mysql problems, since both the previous osgrid failures occurred when both a read and write were attempted in the same second
To use, see the appearance section in opensim.ini.example, set "persist = true", then add the correct connection string for your database.(see mysql-AvatarAppearance.sql in share folder for a example of the table mysql table structure).
This could possible be used in a very small grid, but would mean each region server would need to connect to the same mysql database.
But the work to move the code to one of the grid servers shouldn't be too much.
The ScriptDialogReply packet handler is a bit of a hack job. It is currently handled similar to ChatFromViewer, which will trigger the listen() event, however this is not exactly how LL's implementation works and will/can be fixed up later.
CONFIG SET section key value value value
CONFIG GET section key
CONFIG SAVE (it saves, but does it save correctly?:)
ScriptEngine will react correctly to any config change made while it is running.
Added option to try alternate UDP ports if the one configured is in use.
UDP packets are now bound to the actual outside IP address and hopefully won't "randomly" select IP on multihomed systems.
* You can type 'stats' at the REGION# prompt to get this information in grid or standalone mode
* Don't take these numbers as gospel yet, since for some reason textures displayed from inventory which require downloading from the server are being recorded as assets
rather than textures
* But I don't have any reason to believe they aren't broadly accurate.
* I've put these in so I can tell whether the high memory usage on regions is down to the asset/texture cache
* This will require a prebuild
* DEV: Only adds needed to be implemented since, as far as I can tell, assets cached are currently never released. For my part, seeing large cache memory numbers will
provoke me to think about doing something about this.
* DEV: Now switched to using a singleton to get the stats reporters rather than threading the object through various layers
* DEV: Will refactor the other server stats reporters to do this in one of the next commits
* The hooks are still plugged in too high at the asset server, but then next layer down is the database and this may be refactored soon.
* This change will requires a prebuild
* I intended to think about refactoring the servers to use this common base class, but temporary laziness won out.
* I'm leaving this here as a reminder/mind poke and for reasons of laziness (again)
* If people object to me leaving this kind of thing lying around, please say
* Typing 'stats' on the command line will given total number of successful logins today and yesterday
* A little bit more to come, probably
* Refactoring will follow next
* If you have Debian running, you should get a platform line that says, 'Found Debian!' when starting up your sim.
* If someone running Debian will confirm this does occur, that would be most helpful.
** If the region fails 3 times, then ChildAgentDataUpdates no longer get sent to that region
* Enabled Child_Get_Tasks in grid mode.
* When Child_Get_Tasks is enabled on neighbor regions, the neighbor region uses the client's draw distance to send out prim. This is a lot less likely to flood the client now since the ChildAgentDataUpdate contains both the throttle settings and the draw distance. This means that with this enabled, you can see prim in other regions in grid mode. Very experimental.
* On restart the latest save will be restored rather than the very first dragged in scripts
* Also add previously missed out database commits to separate prim inventory commit path (sigh)
for now. It doesn't build on new mono at the moment, and
isn't currently used. Please make sure this functions on
mono 1.2.6 prior to adding it back in for the main tree.
* Added a Logout message with a name on the console
* Added a *fixme* message to figure out why the current agent session is null
* After updating you may notice that there's a login <user> and also a logout<user>
Possibly solved startup crash.
Moved MySQL Commit() to inside each functions respective dataset access mutex because if not other threads would be waiting until function released mutex and modify the dataset before the function could commit
* This means that you can take an object from a region and rez it somewhere else, with its inventory intact.
* As for earlier, at this stage only scripts can be placed in inventory
* This isn't an efficient implementation, a better one will probably need to come along soonish
* Apologies, any sqlite databases created between r3069 and this change will now contain unfindable assets.
* But sqlite asset databases used previous to r3069 will now work again.
* This change may have been done on purpose, so it's not impossible it will be re-reverted
* Cleaned up copyright notices in AssemblyInfo.cs's
* Added Copyright headers to a bunch of files missing them
* Replaced several common string instances with a static constant to prevent reallocation of the same strings thousands of times. "" -> String.Empty is the first such candidate.
* Added PrimitiveBaseShapeTableMapper to show how it's done
NOTE: Nothing actually works yet - this code should be considered more of educational value until it's all wired together
* Don't forget, you need -loginuri *and* -loginpage
* Ex: -loginpage http://10.1.1.2:8002/?method=login -loginuri http://10.1.1.2:8002/
* The ?method=login is important, don't forget to add it
* If you customize your http_loginform.html file, be sure to keep the form post address as is.
namespace to mirror what is done with MySQL. This will require a
configuration change for people using SQLite for prims which is given
as an example in OpenSim.ini.example
* Added the web_login_key to the users table
* Added happy configurable http error message pages
* This update is large enough to have 'awe' value.. so backup your users or weep.
* Not tested on MSSQL, even though I added code to update the tables!
* I'm keeping it deactivated until some issues are resolved.
* I'm patching it in deactivated so the patch doesn't get outdated
* I've deactivated it by commenting out the handler for the application/xml+llsd content type.
* While I've tested this as much as possible on my setup and found the deactivated code doesn't cause any problems, consider this update experimental (event though it's deactivated)
* Estate managers get abilities like the region owner for now.
* Estate managers, you'll need to request server admin status to be able to activate the estate tools dialog (haven't figured out why this is the case yet)
* Switching from grid mode to standalone or switching grids will make the stored Estate Manager UUIDs not match up with a valid account so you'll see (waiting) listed there instead of a user until you reset them or go back to the grid you added them from.
* You can put UUIDs in those fields to give other users some ability to help manage a sim you own.
* Also started decoding the EstateOwnerMessage packet convolutions.
* The BlockingQueue exposes Contains so we can make sure we don't add a TextureSender to the queue if there's already one present
* introduced some TryGetValue and various code convention stuff
* You can add and remove a friend in grid mode now within the same simulator. It doesn't save yet.
* I got rid of Mr. OpenSim as a friend.. he bothers me /:b...
* This patch addresses a regression where 5 default textures, including the default "new prim" texture, would not be shown unless in viewer cache.
* Further, it supplies a new plywood texture as the base texture, which is purpose-made from locally photo-sourced material (means i took a picture and made it into a tile), which is closer in appearance and color to the Linden one.
* It changes the default new prim texture UUID to match the UUID generated by the viewer when "default" is clicked, which is a texture UUID hardcoded in some scripts as well, so it would cause compatibility issues if not addressed.
* The patch fixes a number of UUID duplications (same UUID for asset and inventory items) and capitalization issues in the XML files.
* It also includes new versions of 4 other textures, which were no longer loadable in grid mode.
* Charles, this will fix the red issue.
* Same situation, the first run updates the tables (and gives you a ton of red errors), the second run and everything works as expected.
are not displaying (which is why default cubes are coming up as grey plywood unless you happen to have that texture cached). These
were working before so they must have broken in the last month. Might be something to do with the fact that these identify (using file
under linux) as jpeg2000 files, while all the other working textures identify simply as data.
* PrimitiveBaseShape: The textures are now exposed as a 'TextureEntry Textures'; all serialization still using the 'byte[] TextureEntry' for backwards compatibility.
* Scene: Re-added AddTree, since the Tree type isn't gone from libsl, merely relocated.
* SAVE YOUR PRIM DATA, THIS MAKES CHANGES TO YOUR PRIMS TABLE
* The first time you run OpenSim after updating past this revision, you'll see a lot of Errors. Be calm, shutdown the simulator, and start it again and your prims table will be updated.
* MSSQL added the fields to the Initial CreateTable section, however, you'll need to add the fields to your prims table if you want it to persist.
This may not be the best solution in the long run, but should improve things for now.
This may also improve reliability when updating inventory item metadata (e.g. renaming an item) and in retrieving textures
for the main map view.
* This fixes a *bunch* of Mantis bugs related to the following
* -- Neighbouring simulators not appearing after relog
* -- Login to simulator only to be logged off by simulator.
* -- ThreadAbort
* -- Unable to shutdown circuitCode: x
Inventory contents retrieval and persistent region storage standalone now appear to work as well as they did before :)
This patch will not fix grid problems.
May be bugs present due to conversions I didn't spot.
I personally probably don't have any more time for this today. I'm also not entirely convinced this is the right way forward
so this might be a handy pause for thought. I'll also be delighted if I wake up tommorrow and everything is fine again.
Apply method to UUID crud in SqliteInventoryStore as an initial test
This appears now to successfully recover inventory upon login
This will almost certainly only work on standalone
* added Util.Clip(value, min, max)
* modified asset cache's numPackets calculation to use max packet size (600) instead of 1000
* removed a few magic numbers
* Added Incomplete Handling of the EstateCovenantRequest packet
* Added Incomplete Handling of the EstateOwnerMessageRequest.ChangeEstateCovenantid method
* Fixed a race condition with avatar animations
* I've fundamentally changed a few things, so this is experimental
* The routine that I used needs to be tested on Linux. I don't expect it to cause a problem, but hey, it might.
* Child agents are still not logged off properly, so when the first set time out, the second set get logged off also, on the second log in if the second login is initiated before the first one fully times out.
* Now sending South and East cords to the terrain editor..
* No new functionality from a user perspective
* Programming wise, there's enough information to get the select based terrain editor working in an upcoming revision.
* The AssetServerBase is now responsible for dequeueing, the server implementations merely recieves ProcessRequest( AssetRequest req )
* Catchall added around queue processing thread so thread won't abort on exceptions.
* Fixes weirdness when typing and sitting at the same time
* Should fix bug #32 (getting stuck in edit appearance pose)
* Crouchwalk and possibly jump may need more looking into
Refractored the TextureDownloadModule (but currently to make debugging easier, it is running as a non shared module, so this results in a instance of this module being created for each region (and a extra thread per region), this will be changed back soon.
Removed the old texture handling/sending code from AssetCache.
A few other small changes/fixes.
* Added List<RegionInfo> m_neighbours to Scene
* Hooked up the OnRegionUp event to m_neighbours list
* Modified RegionInfo to have a bool commFailTF value so that we can skip neighbors that fail. (when the region comes up, this gets reset to false and the region will try again.
* Added SetChildAgentThrottle(byte[]) to IClientAPI
* Several other insignificant changes related to passing child pertanant agent data from sim to sim.
This corrects an off by one mistake in my previous patch - with this patch
the user root inventory folder should once again be correctly returned
under sqlite.
This was done correctly for mysql and mssql, my only (feeble) defence is
that with sqlite the code was a little different and I was eliminating
clauses rather than slightly modifying them.
* Will be using this glue Tomorrow/today to tell other regions what the agent's draw distance is and what region they're actually in so the region can make decisions on what prim to send, if any.
* BACKUP YOUR PRIM BEFORE UPDATING TO THIS and then double check the prim permissions after applying it with a different avatar (then the master avatar or the prim owner avatar).
* Also, beware that any objects created under the old permission scheme may react oddly. They may automatically allow anyone to modify them, (which you'll then have to un-set).
* It's hacked support because when 'anyone can move is set', any avatar can modify the prim (texture, shape, scale, etc)
me on Mono 1.2.4, which led to client crashes. I think the Timer.Stop()
wasn't doing what was desired on Mono. The Queue refactoring should address
the readability issues lbsa71 was working on as soon as I get the merge together.
* Un-hackerized generating the client_flags
* Now handling the ObjectPermissions Update packet
* Warning: Backup your prim before updating. If you fail to do so and something goes wrong then, All Yr prim are belong to us!
* To test, click the communicate button, then click the 'add' button and type a name. You will get a list of closely matching avatar names.
* Before this'll work on your sim in grid mode, the UGA needs to be updated. If it isn't updated, then you'll get a message on the sim console when searching like, 'got invalid queryid'.
* Added SSL Option to User Server to allow logins to be done via SSL.
* Added sane handling for when Remote Admin Plugin configuration is not found
* Added some performance boosts to an area of libTerrain which was highlighted in profiling.
While exploring what it would take to get the 'new script' button working,
I encountered the fact, some way down in the rabbit hole, that if a user
renamed an item in their inventory and logged out (without a restart of
the simulator), on log in the new name was not preserved.
As far as I can see, this was because any updates which didn't occur
inside a transaction were ignored by opensim. This patch pays attention
to those changes. It generates a new asset when an item is updated and
changes the user's inventory properties appropriately. I believe this
behaviour is in line with the copy-on-write semantics used in the Second
Life protocol - perhaps it could be optimized if we knew for sure that the
only copy of the object was in the user's inventory.
This also means that if you rename an item (e.g. a script) before you drag
it into an object's inventory, the inventory will receive the item's most
recent name and description.
To use , both the user server and Inventory server need to be running this latest revision. (older regions should be able to still be used, just the user won't have inventory on them). Also and HERE IS THE BIG BREAK ISSUE, currently, so that the initial inventory details for a user are added to the inventory db , you need to recreate the accounts using the user server "create user" feature. It should be quite easy to manual populate the inventory database instead but I someone else will need to look into that) Also I've only tested using SQLite as the database provider, there is a Mysql inventory provider but I don't know if it works (SQLite is set as default, so you will need to change it in the inventory server config.xml)
Some work towards persisting Avatar Appearance (what is being worn).
Added OnAvatarNowWearing event to IClientAPI that is triggered by AgentIsNowWearing packets.
stub code to subscribe to this event in AvatarFactoryModule.
Todo: code needs to be added to AvatarFactoryModule to save the uuids to a database and then read them back when that modules TryGetIntialAvatarAppearance() method is called.
Done some changes to Scene to make it easier to subclass it: including changed some private fields to protected and made some methods virtual.
* This helps people who have multiple records in their useragent table.
* This update should be looked at.. because really what it should be doing is initiating a grid wide logoff.
Deleted the CreateTerrainTextureInitial Method which was a 99% duplicate of CreateTerrainTexture, with just a bool field setting difference. That bool is now passed to CreateTerrainTexture as a param.
* The sims properly restart, however they don't yet notify the existing avatars that they are up. To see the sim again, you'll need to log-out and back in until I can figure out how to get the proper data to the sims and to the avatar so they reconnect again.
* Added the child_get_tasks OpenSim.ini flag for testing the UDP packet sending code and packet throttler. This flag gets purposely disabled in grid mode. This flag also has the consequence that you can see the prim in neighboring regions without going into them. Be warned, this causes tons of dropped packets.
MW